Petri
Open-source auditing tool where auditor agents probe target models in simulated scenarios and judges score misaligned behaviors.
Anthropic cards use Petri to compare Claude with other labs' models.
- Made by
- Anthropic
- Released
- October 2025
- Kind
- Deception and alignment
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- All public
- Last in a system card
- July 2026
- Last model added
- July 2026
8
System cards report it, from 2 labs
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for Petri yet.
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
8 system cards report it, from 2 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Jul 9, 2026 |  Muse Spark 1.1 system card | 1.74 | Petri 3.0 overall misaligned behavior, mean judge score, averaged over three auditors (Table 16) | Alignment |
| May 28, 2026 | Claude Opus 4.8 system card | Petri 3.0 run by Meridian and UK AISI, 346 seeds, six metrics (misaligned behavior, verbalized evaluation awareness, cooperation with human misuse, user deception, sycophancy, encouragement of user delusion), chart without printed values (6.2.3.3, Figure 6.2.3.3.A) | Alignment | |
| Apr 16, 2026 | Claude Opus 4.7 system card | Petri 2.0, six metrics including misaligned behavior, 362 investigations per model, charts without printed values (6.2.3.3, Figure 6.2.3.3.A) | Alignment | |
| Apr 7, 2026 | Claude Mythos Preview system card | Petri 2.0 external comparison on a significantly earlier snapshot, 362 investigations, six metrics including misaligned behavior, charts without printed values (4.2.3.2, Figure 4.2.3.2.A) | Alignment | |
| Feb 17, 2026 | Claude Sonnet 4.6 system card | Petri 2.0, six metrics including misaligned behavior, charts without printed values; verbalized evaluation awareness 1.2 of 10 stated in text (Figure 4.5.3.A) | Alignment | |
| Feb 5, 2026 | Claude Opus 4.6 system card | Petri 2.0 external comparison, 362 investigations, six metrics including misaligned behavior and verbalized evaluation awareness, charts without printed values (Figure 6.2.5.5.A) | Alignment | |
| Nov 24, 2025 | Claude Opus 4.5 system card | Petri external comparison, pre-final Claude Opus 4.5 Preview snapshot, 362 investigations, concerning, evaluation awareness, misuse cooperation, user deception and sycophancy, chart without printed values (Figure 6.2.4.A) | Alignment | |
| Sep 29, 2025 | Claude Sonnet 4.5 system card | Pre-release open-source automated behavioral auditor, 111 seed instructions, auditors and graders from several developers, misaligned behavior score chart without printed values (7.1.3.3, Figure 7.1.3.3.A) | Alignment |
Versions
More in Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6% - Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts.
Gemini 4 Carbon 0% - DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics. Muse Spark 1.2 88.4%
- Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
- IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%